The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes session transcripts and logs to identify friction and patterns. This ingestion of data from previous interactions creates a surface for indirect prompt injection, where malicious instructions present in the history could influence the agent's behavior during the rule generation process.
Ingestion points: The skill reads local machine logs and the current session's transcript (SKILL.md, Step 1).
Boundary markers: The instructions do not define specific delimiters or warnings to ignore instructions embedded in the ingested logs.
Capability inventory: The agent is instructed to write changes to repository files, update documentation (e.g., AGENTS.md, CODING_STANDARDS.md), and modify project skills in .agents/skills/ (SKILL.md, Step 4 and "Where things go" section).
Sanitization: No explicit sanitization or filtering of the log content is described before processing.
[COMMAND_EXECUTION]: The skill performs shell operations to locate its configuration and finalize updates within the local repository environment.
Locates its global source repository using find ~/code ~/src ~/projects -maxdepth 3 -type d -name agent-skills, readlink, and git -C <path> remote get-url origin.
Executes build commands such as make or make rules within the target repository to regenerate files after approved modifications have been written (SKILL.md, Step 4).

durable-rules